On the Parameter Identifiability of Partially Observed Linear Causal Models